conf.yaml

experiment:
  wildtype_AA_seq: "MSTETLRLQKARATEEGLAFETPGGLTRALRDGCFLLAVPPGFDTTPGVTLCREFFRPVEQGGESTRAYRGFRDLDGVYFDREHFQTEHVLIDGPGRERHFPPELRRMAEHMHELARHVLRTVLTELGVARELWSEVTGGAVDGRGTEWFAANHYRSERDRLGCAPHKDTGFVTVLYIEEGGLEAATGGSWTPVDPVPGCFVVNFGGAFELLTSGLDRPVRALLHRVRQCAPRPESADRFSFAAFVNPPPTGDLYRVGADGTATVARSTEDFLRDFNERTWGDGYADFGIAPPEPAGVAEDGVRA"
  ligand_smiles: "CCCc1ccc(O)c(OC)c1"
  device: "cuda"
  render_timesteps: 20
  time_limit_h: 4
  sequence_tracker_max_length: 1000
  number_parallel_envs: 4
on_policy_trainer:
  epochs: 1000000 
  batch_size: 100
  steps_per_epoch: 20
  repeat_per_collect: 1
  steps_per_collect: 100
  replayBuffer:
    total_size: 10000
agents:
  picker_ppo:
    gumbel_dist_temperature: 1
    gumbel_k: 3
    policy:
      eps: 0.2 # 0.2
      value_clip: False # True
      advantage_normalization: True
      recompute_advantage: False
      vf_coef: 0.5
      ent_coef: 0.1 # 0.01 -> 0.1
      gae_lambda: 0.95
      discount_factor: 0.99
      reward_normalization: False # Keep 'False'!
    adam:
      learning_rate: 0.0001 #1e-4
      epsilon: 1e-5
  filler_plm:
    evodiff_model_size_parameters: 640
hydra:
  run:
    dir: ./experiments/${now:%Y-%m-%d}/${now:%H-%M-%S}
  job:
    chdir: True
  verbose: False